Analisis Pengaruh Harga Emas Dan Indeks S&P 500 Terhadap Peramalan Harga Bitcoin Menggunakan Model Deep Learning CNN-LSTM

Pradipta, I Gusti Ngurah Adhya (2026) Analisis Pengaruh Harga Emas Dan Indeks S&P 500 Terhadap Peramalan Harga Bitcoin Menggunakan Model Deep Learning CNN-LSTM. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5026221184-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (3MB) | Request a copy

Abstract

Volatilitas harga Bitcoin yang ekstrem menjadi tantangan utama dalam pengambilan keputusan investasi. Untuk meningkatkan akurasi prediksi di tengah ketidakpastian pasar, penelitian ini mengusulkan model hibrida Convolutional Neural Network (CNN) dan Long Short-Term Memory (LSTM). Model ini menggunakan pendekatan multivariat dengan mengintegrasikan harga emas dan indeks S&P 500. Pengujian dilakukan menggunakan data harian periode 2020 - 2024. Dampak penambahan variabel eksternal tersebut diuji melalui empat skenario eksperimen (studi ablasi) dan dievaluasi menggunakan metrik Root Mean Squared Error (RMSE) serta Mean Absolute Percentage Error (MAPE). Hasil optimasi menghasilkan konfigurasi hiperparameter terbaik yang meliputi 32 filter, ukuran kernel 2, fungsi aktivasi Tanh, 50 unit LSTM, tingkat dropout 0.2, serta penggunaan optimizer Adam dengan learning rate sebesar 0.01. Berdasarkan studi ablasi, Skenario C (penggabungan riwayat harga Bitcoin dan emas) menghasilkan performa prediksi paling presisi dengan nilai kesalahan MAPE terendah sebesar 2.55%, mengalahkan skenario univariat dengan MAPE 2.60%. Hasil eksperimen mengonfirmasi bahwa emas memberikan kontribusi positif dalam menangkap momentum pembalikan harga, sedangkan penambahan indeks S&P 500 justru menimbulkan derau informasi yang menurunkan akurasi model. Saat diaplikasikan untuk peramalan 14 hari dan 30 hari ke depan, model secara akurat memproyeksikan terjadinya fase koreksi turun pasca tren kenaikan harga. Penelitian ini membuktikan bahwa model CNN-LSTM dengan seleksi variabel dan hiperparameter yang tepat dapat menjadi alat bantu prediksi yang andal bagi investor di pasar kripto.
===================================================================================================================================
The extreme price volatility of Bitcoin poses a major challenge in investment decision-making. To improve prediction accuracy amidst market uncertainty, this study proposes a hybrid Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) model. This model employs a multivariate approach by integrating gold prices and the S&P 500 index. The experiment was conducted using daily data from the 2020 - 2024 period. The impact of incorporating these external variables was tested through four experimental scenarios (ablation study) and evaluated using the Root Mean Squared Error (RMSE) and Mean Absolute Percentage Error (MAPE) metrics. The optimization yielded the best hyperparameter configuration comprising 32 filters, a kernel size of 2, Tanh activation function, 50 LSTM units, a dropout rate of 0.2, and the use of the Adam optimizer with a learning rate of 0.01. Based on the ablation study, Scenario C (the combination of Bitcoin price history and gold) produced the most precise prediction performance with the lowest MAPE value of 2.55%, outperforming the univariate scenario which yielded a MAPE of 2.60%. The experimental results confirmed that gold provides a positive contribution in capturing price reversal momentum, whereas the addition of the S&P 500 index introduced information noise that reduced the model's accuracy. When applied to 14-day and 30-day forecasting, the model accurately projected a downward correction phase following an uptrend. This research proves that the CNN-LSTM model, with proper variable selection and hyperparameters, can serve as a reliable prediction tool for investors in the cryptocurrency market.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Bitcoin, CNN-LSTM, Peramalan Multivariat, Deep Learning, Bitcoin, CNN-LSTM, Multivariate Forecasting, Deep Learning
Subjects:	H Social Sciences > HA Statistics > HA30.3 Time-series analysis T Technology > T Technology (General) > T174 Technological forecasting
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Information System > 57201-(S1) Undergraduate Thesis
Depositing User:	I Gusti Ngurah Adhya Pradipta
Date Deposited:	24 Jul 2026 02:05
Last Modified:	24 Jul 2026 02:05
URI:	http://repository.its.ac.id/id/eprint/137013

Actions (login required)

View Item